WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=GCse4OMzY8U

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/GCse4OMzY8U.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.760 --> 00:00:13.880
Um hey everyone.

s3
00:00:13.880 --> 00:00:23.280
My name is Afaan and I'm I'm a lead for compute trading at Jump Trading, which is a proprietary trading firm, alongside my co-lead over here, Alejandro.

s4
00:00:23.520 --> 00:00:34.800
So, today we're going to be looking over some of the market structure as it applies to compute trading, what compute trading even is, um what we can see it becoming.

s5
00:00:34.800 --> 00:00:43.320
And I think we'll start with some macro, look at micro structure, and then talk a little bit about where we see things in policy.

s6
00:00:44.520 --> 00:00:52.320
So, the title of the talk is the dark pools of compute, but some of you guys are in finance, so you might be confused as to why I'm using this terminology.

s7
00:00:52.320 --> 00:00:54.200
Uh what do I mean by dark pool here?

s8
00:00:54.200 --> 00:00:57.160
Essentially, what I'm referring to is an unlit market.

s9
00:00:57.160 --> 00:01:02.920
Um right now, most compute trades in blocks, physical compute here, and these are private quotes.

s10
00:01:02.920 --> 00:01:06.120
We have a lot of brokers and these are long bilateral contracts.

s11
00:01:06.120 --> 00:01:09.680
Now, at the end of the day, these are really commercial agreements.

s12
00:01:09.680 --> 00:01:14.760
So, we're using the term trading here to kind of like look at how it's transacted between parties.

s13
00:01:14.760 --> 00:01:28.040
These commercial agreements tend to be some mixture of an MSA, an SLA, and an order form, cluster specific uh with hardware specific language in each kind of um in each order.

s14
00:01:28.040 --> 00:01:35.720
And but when we're looking at pricing and looking at some of these other kind of uh metrics, a lot of the time, most of the talk is around the skew.

s15
00:01:35.720 --> 00:01:37.760
So, what the chip itself is, the base board.

s16
00:01:37.760 --> 00:01:47.720
So, H100, H200, B300, uh etc. And so, um a lot of the negotiation here happens kind of outside the public view.

s17
00:01:47.720 --> 00:01:57.280
There's no, you know, like exchange for physically traded compute that's widely, you know, adopted and you know, used for everyone goes there to like take a look and trade

s18
00:01:57.280 --> 00:01:58.240
uh compute.

s19
00:01:58.240 --> 00:02:00.960
These are done in chats, right?

s20
00:02:00.960 --> 00:02:02.520
They're out of the public view.

s21
00:02:02.520 --> 00:02:07.840
Um there's public marketplaces, there's price references, there's a lot of um you know, on-demand marketplaces.

s22
00:02:07.840 --> 00:02:18.760
However, for these actual tenured, so you know, on a given term for a fixed amount of capacity uh with a certain total contract value, these trade over Slack messages,

s23
00:02:18.760 --> 00:02:21.120
X messages, Telegram group chats.

s24
00:02:21.120 --> 00:02:26.800
It's a bunch of OTC kind of channels that are all over in these kind of I want to call them like neo neo chat.

s25
00:02:26.800 --> 00:02:39.920
Um and some of you might be familiar with commodities and how a lot of you know, there's chatting happening on Ice Chat, Bloomberg Chat, etc. This is kind of where where the where the kind of markets here lie.

s26
00:02:39.920 --> 00:02:45.800
And so I want to look at a little bit here on the macro market structure and a little bit on the players here as well.

s27
00:02:45.800 --> 00:02:50.760
And so this was a recent kind of a16z, some of you guys might have seen this.

s28
00:02:50.760 --> 00:02:56.959
So I'm I'm putting this out there because you know, you can go on a16z's actual X and take a look at this graph yourself.

s29
00:02:56.959 --> 00:03:01.920
What we're looking at is that the cost of a given capability has fallen sharply.

s30
00:03:01.920 --> 00:03:10.840
The bill of course still depends on the model usage, capacity, and other factors such as speed is a big thing right now with the new code ChatGPT models.

s31
00:03:10.840 --> 00:03:16.080
Um cheaper tokens unlock more usage and a team can do more work for the same budget.

s32
00:03:16.080 --> 00:03:20.480
So what we're seeing is that a lot of teams are simply just using way more tokens.

s33
00:03:20.480 --> 00:03:27.440
So we have this latent demand that hasn't really been hit yet and people were just going to use more with the same budget that they were using before.

s34
00:03:27.440 --> 00:03:31.480
One model's cost more and they're just expending more tokens.

s35
00:03:31.480 --> 00:03:34.880
Um so applications that were too expensive become possible.

s36
00:03:34.880 --> 00:03:43.519
Um however, as we're kind of talking about compute here, the capacity underneath still has to be built and paid for um even as the cost of a particular task comes down.

s37
00:03:43.519 --> 00:03:51.200
And so the important distinction is between the price of the intelligence, the number of tasks people want to run, and the amount of hardware needed to serve them.

s38
00:03:53.560 --> 00:04:00.440
And so this is where we get to kind of the Jevons paradox and a little of some of the stuff that we've been looking at lately where we had

s39
00:04:00.440 --> 00:04:11.320
a very interesting thing where, you know, intuitively you'd assume that compute and inference kind of like rises together, correlates very, you know, very closely as you can see.

s40
00:04:11.320 --> 00:04:17.519
You can see there's a little bit of a decoupling that happened and token prices declined further but compute prices have remained robust.

s41
00:04:17.519 --> 00:04:19.280
And so like what what can be an explanation for this?

s42
00:04:19.280 --> 00:04:23.000
Now, there's a lot of like potential explanations.

s43
00:04:23.000 --> 00:04:32.800
One here is the whole Jevons paradox is that even though token prices and inference is getting extremely efficient to serve on these new baseboards and and and and GPUs,

s44
00:04:32.800 --> 00:04:34.800
people are just consuming a lot more.

s45
00:04:34.800 --> 00:04:47.680
And so the demand for compute rises and people just consume a lot more tokens even if the cost per cost per million for input and for caching and, you know, for output becomes a lot more efficient.

s46
00:04:47.680 --> 00:04:52.120
And so making the task cheaper, encouraging the use that way resource saving per task.

s47
00:04:52.120 --> 00:04:58.800
For inference clouds, that's going to be more applications, serving more tokens, bigger market share.

s48
00:04:58.800 --> 00:05:02.560
And for app layer individuals, that means you guys can serve to a lot more kind of customers.

s49
00:05:02.560 --> 00:05:07.040
So anyone who's building utilizing the token infrastructure.

s50
00:05:08.240 --> 00:05:11.840
And so what really goes into, you know, the macro structure.

s51
00:05:11.840 --> 00:05:17.040
Now like this is very, you know, bare bones here but you've got your chips which we've been talking about.

s52
00:05:17.040 --> 00:05:26.520
You have power, you have buildings, grid connections and then of course you need a lot of capital to get these build outs going and then that's kind of been the whole talk here.

s53
00:05:26.520 --> 00:05:35.240
And so um chips, powers, buildings, money, they all have to come together before, you know, you can even get to an hour of like actual usable GPU per hour compute.

s54
00:05:35.240 --> 00:05:45.360
And so the IEA April update projects that the global global data center electricity is going to rise from 485 terawatt hours in 2025 to 915 2030.

s55
00:05:45.360 --> 00:05:51.040
And so a lot of the talk right now is around build-outs, about getting the power grid set up.

s56
00:05:51.040 --> 00:05:52.920
Will we have enough power grid here?

s57
00:05:52.920 --> 00:05:56.160
Is there other energy sources that we can tap into?

s58
00:05:56.160 --> 00:05:59.720
Where are we, you know, able to get the land to actually be able to do this?

s59
00:05:59.720 --> 00:06:06.160
And so, colocation sites and data center operators, and people who are doing the financing the build-outs or the lenders themselves, are all kind of looking into this.

s60
00:06:06.160 --> 00:06:16.560
And that type of everything that's upstream, essentially, of the compute is going to then flow down into the compute uh costs and also usage.

s61
00:06:16.560 --> 00:06:27.320
And so, what does this like actual macro map look like underneath a quote chip and system vendors supply the GPU servers and networking, sites and operators supply the power,

s62
00:06:27.320 --> 00:06:40.080
space assembly and management right here, GPUs memory servers and networking, um capacity providers, these are hyperscalers and neo clouds and any other operator that's like operating similar to how a neo cloud would operate, reselling bare metal capacity,

s63
00:06:40.080 --> 00:06:40.880
and then the users.

s64
00:06:40.880 --> 00:06:42.040
So, who's the actual end user?

s65
00:06:42.040 --> 00:06:52.280
The startups, the inference clouds, uh labs, neo labs, and also anyone on the app layer and enterprises for anyone doing open weight stuff as well.

s66
00:06:52.280 --> 00:07:00.800
And um these boxes, like I've kind of boxed them out here, but, you know, there's people can be all of the above.

s67
00:07:00.800 --> 00:07:06.360
A chip and system vendor, like someone could do a full end-to-end build-out and also be the end user of everything.

s68
00:07:06.360 --> 00:07:12.800
And so, there's a little bit of um variation that that that occurs here.

s69
00:07:17.960 --> 00:07:27.440
And so, as I was kind of mentioning about the inference clouds before, they might rent a cluster and they'll serve tokens off of it and they'll use that revenue to the pay off the cluster.

s70
00:07:27.440 --> 00:07:32.360
Um and so, that might be a use case for the actual inference cloud themselves.

s71
00:07:32.360 --> 00:07:41.320
However, for hyperscalers, frontier labs, and larger neo clouds, they can make substantial commitments because they have the ability to get into the large equity, they have the ability to take on more debt.

s72
00:07:41.320 --> 00:07:46.000
And uh they have the ability to do long horizon kind of deals.

s73
00:07:46.000 --> 00:07:50.800
So, 10-year deal opening I contracted it out, etc. So, of that nature.

s74
00:07:50.800 --> 00:07:54.040
However, anyone in the middle market isn't actually able to do this.

s75
00:07:54.040 --> 00:08:02.640
Who's underwriting a Series B startup with enough capacity to get in a 200 million plus contract over 3 years for compute?

s76
00:08:02.640 --> 00:08:05.200
How many equity rounds will they have to raise, right?

s77
00:08:05.200 --> 00:08:14.280
Will they have to use their entire equity round to fund just the actual cluster and then they don't have any remaining capital left to use for actual costs?

s78
00:08:14.760 --> 00:08:18.280
And so, we're looking at a little bit of the financing chain here and how that plays in.

s79
00:08:18.280 --> 00:08:23.960
So, lenders are financing the buildout against contracted revenue even in many cases for inference clouds.

s80
00:08:23.960 --> 00:08:25.360
And so, they want to finance the buildout.

s81
00:08:25.360 --> 00:08:27.800
They want confidence that their loan will get repaid.

s82
00:08:27.800 --> 00:08:32.880
I mean, like how are you going to underwrite a startup like a private company that doesn't have like much else going for them right now.

s83
00:08:32.880 --> 00:08:34.400
Like what are the, you know, they're not publicly traded.

s84
00:08:34.400 --> 00:08:36.560
There's not that much cash flows, balance sheet, income statement.

s85
00:08:36.560 --> 00:08:45.040
They might even be pre-profit, pre-revenue, etc. And so, what they're looking at is that can the person get contracted revenue over a certain scale.

s86
00:08:45.040 --> 00:08:51.400
And so, Coreweave's public filings give large-scale example infrastructure financing supported by take or pay customer contracts.

s87
00:08:51.400 --> 00:08:56.440
And so, this diagram kind of shows the smaller illustrative example here.

s88
00:08:56.440 --> 00:09:05.760
And the mismatch is that hardware financing they want very long terms, but customers are only able to afford or don't really want to like get into a long-term financing over 36 months.

s89
00:09:05.760 --> 00:09:17.720
They have to put up on up to 15 to 25% is what Coreweave has quoted that they've seen their median prepayment to be, 15 to 25% up front on a 100 million TCV cluster.

s90
00:09:17.720 --> 00:09:21.920
That's 15 to 25 million dollars that a startup has to put up front.

s91
00:09:21.920 --> 00:09:24.640
That might be their entire round.

s92
00:09:25.640 --> 00:09:29.000
And so, this is kind of a look at what I'm you know, referencing here.

s93
00:09:29.000 --> 00:09:32.720
It's like you're raising 40 million, you need a 100 million contract over 3 years.

s94
00:09:32.720 --> 00:09:34.200
You put down 30% prepayment.

s95
00:09:34.200 --> 00:09:35.480
This is an illustrative example.

s96
00:09:35.480 --> 00:09:41.360
You've got 10 10 million left to run the business, um and so you might have this kind of step function as you can see here.

s97
00:09:41.360 --> 00:09:47.280
So, that's your contracted capacity and what you can actually in the demand is like, you know, uh below that.

s98
00:09:47.280 --> 00:09:54.920
And so, now every time that you have to get into a new cluster, it's like tied to your equity round or whatever else it may be, but you might have to cap out and turn away customers.

s99
00:09:54.920 --> 00:09:55.960
So, what are your options?

s100
00:09:55.960 --> 00:10:10.520
You either turn away customers until you can get into your next, um you know, equity round so you can put down another 30% prepayment so you can get next little bit of capacity going, or do you use on demand to close the area under the gap and bridge the shortfall?

s101
00:10:10.600 --> 00:10:12.000
That's like an That's like an option now.

s102
00:10:12.000 --> 00:10:13.480
Now you have to look into the economics.

s103
00:10:13.480 --> 00:10:14.080
What does that look like?

s104
00:10:14.080 --> 00:10:22.640
What's the difference between, you know, maybe instead what they do is they go to a lender to finance the prepayment component, but if the yield on the prepayment component that they're getting financed

s105
00:10:22.640 --> 00:10:32.400
is less than, um them just buying it out on demand at that time if they're able to get into that amount of capacity, then does the economics make sense there?

s106
00:10:36.480 --> 00:10:39.120
And so, here are some of the operating models that we've seen.

s107
00:10:39.120 --> 00:10:42.280
It's like you can build and operate the full stack.

s108
00:10:42.280 --> 00:10:47.560
Hardware, site, your team is going to arrange power, permit, the assembly, the management.

s109
00:10:47.560 --> 00:10:52.839
You can also own servers in a colocation facility with more or less the assembly and operation handled by a partner.

s110
00:10:52.839 --> 00:11:06.760
Um on the other end, an off-take contract buys defined capacity and the provider is going to own and operate the asset, and they might not give you complete root access to the bare metal or there might be kind of like levels to how much access you get depending on if it's tenant, multi-tenant,

s111
00:11:06.760 --> 00:11:10.839
and etc. type of like configurations that can also affect them the pricing.

s112
00:11:10.839 --> 00:11:23.040
Um And so, some businesses are going to value the physical access, some businesses are going to value, um you know, confidential compute, uh you know, they might value certain topology changes or the ability to reconfigure the system.

s113
00:11:23.040 --> 00:11:36.000
And so, if that is something for your business and they might actually want to have more and more control and build and operate the stack and the economics for them might make sense given the opportunity cost they would have to give up um if they were to just take rental off take contracts.

s114
00:11:36.000 --> 00:11:39.120
So they might be willing to pay more there.

s115
00:11:44.520 --> 00:11:48.440
And so here's how I ownership and rental is a very like crude graph.

s116
00:11:48.440 --> 00:11:52.240
How ownership and rental can compare as utilization arises.

s117
00:11:52.240 --> 00:12:02.680
And so what we're looking at is that usually you know an off take contract or these off take contracts are getting to the cost of actually almost building out and financing your own cluster.

s118
00:12:02.680 --> 00:12:15.360
And it's not because I mean if you're spending a hundred million over a three year period then maybe if the economics make sense it would have made more sense to simply buy out the compute operator yourself outside of the time cost and then off the operational cost that might exist.

s119
00:12:15.360 --> 00:12:28.920
So there's a there's a equilibrium that exists that that has to be followed where you know the ownership cost outweigh the you know rental cost and the rental cost

s120
00:12:28.920 --> 00:12:30.600
should sit below the ownership cost.

s121
00:12:30.600 --> 00:12:39.040
But if you're utilizing at a such a rate and you know in such variable kind of increments it might be best to just own the stack out right.

s122
00:12:42.560 --> 00:12:50.400
And so we're seeing this new dynamic that compute the compute procurement affects how much runway a growing company actually has.

s123
00:12:50.400 --> 00:12:53.560
And so this also gives investors an interest in computer procurement.

s124
00:12:53.560 --> 00:12:54.920
So we're kind of seeing this.

s125
00:12:54.920 --> 00:13:00.040
Access and payment terms affect how much runway their port codes are going to have from a VC perspective.

s126
00:13:00.040 --> 00:13:05.600
So helping a company buy a suitable capacity can improve the underlying investment and they're very connected.

s127
00:13:05.600 --> 00:13:05.800
Right?

s128
00:13:05.800 --> 00:13:12.240
So they're able to procure and they're able to actually find compute capacity for their port codes and and and fund that.

s129
00:13:12.240 --> 00:13:24.839
And so that also brings another interesting aspect to the macro structure into the micro structure is that who helps place the capacity and how does this offer get compared and what risks these contracts can can transfer.

s130
00:13:26.960 --> 00:13:32.280
So we have a lot of different players around this GPU hour that are helping kind of shape this this market here.

s131
00:13:32.280 --> 00:13:34.160
You've got your price discovery individuals.

s132
00:13:34.160 --> 00:13:46.200
So, people with indexes, data, the benchmarks that we're all looking at right now and referencing, which would be used to, you know, in the future launch the futures based on the benchmark rates,

s133
00:13:46.200 --> 00:13:53.640
you know, talked about in in the associated contracts per spec per pending approval.

s134
00:13:53.640 --> 00:13:56.360
And then there's the physical matching venue auctions aggregators.

s135
00:13:56.360 --> 00:13:58.160
So, anyone who can match capacity.

s136
00:13:58.160 --> 00:14:03.160
We're looking at essentially decentralized marketplaces for on-demand compute.

s137
00:14:03.160 --> 00:14:11.160
You're looking at liquid pools of compute that offer like liquid pool services and parted out themselves and have their own standardized contract.

s138
00:14:11.160 --> 00:14:12.440
You're looking at bare metal resellers.

s139
00:14:12.440 --> 00:14:16.720
You're looking at neo clouds themselves doing larger capacity sales.

s140
00:14:16.720 --> 00:14:21.320
And then you're looking at individual brokers and individual operators here as well on the bare metal reselling.

s141
00:14:21.320 --> 00:14:23.280
You have desk and brokers.

s142
00:14:23.280 --> 00:14:25.280
They're looking at the bilateral quotes and execution.

s143
00:14:25.280 --> 00:14:27.240
They might know a lot of people.

s144
00:14:27.240 --> 00:14:28.360
They might be able to move fast.

s145
00:14:28.360 --> 00:14:35.720
They might be able to see quotes at very similar specs in one place and know that there's a buyer that is extremely inelastic to price right now.

s146
00:14:35.720 --> 00:14:40.920
It needs the compute immediately and they can get the RFS in time so they can charge a decent spread on it.

s147
00:14:40.920 --> 00:14:45.560
And then what they do is they're going to have the source negotiating and then they're going to be carrying and placing the risk.

s148
00:14:45.560 --> 00:14:49.600
They're going to have to operate that cluster for the person and it has to be worth that spread.

s149
00:14:49.600 --> 00:14:56.720
And you know, you have looked in financial markets, the exchanges that are coming that are based on the GPU hour, the futures and derivatives.

s150
00:14:56.720 --> 00:15:01.760
People are going to be using it to take a look at how to maybe hedge out and look into more financial structures.

s151
00:15:01.760 --> 00:15:02.960
So, hedge out risk.

s152
00:15:02.960 --> 00:15:04.720
So, hedge out, you know, my balance sheet risk.

s153
00:15:04.720 --> 00:15:07.800
Maybe get into more interesting agreements.

s154
00:15:07.800 --> 00:15:13.280
We're looking we're seeing kind of agreements such as prepayment payment facilities are coming online.

s155
00:15:13.280 --> 00:15:25.320
We're seeing people utilize futures and are interested in utilizing futures and other kind of um um options to be able to There's demand for this to be able to hedge out the risk that they currently have on the books.

s156
00:15:25.320 --> 00:15:38.160
We have buy side individuals looking for flex contracts, flexible contracts like hey, is there ability that if I lock myself into a 3 to 5 year contract that I'll be able to you know, at least upgrade when a new chip that comes out or maybe I'll be able to roll it over or maybe let me sell it

s157
00:15:38.160 --> 00:15:40.080
back to you.

s158
00:15:40.080 --> 00:15:41.360
And then there's a credit financing.

s159
00:15:41.360 --> 00:15:45.760
So equipment lenders, leasers, specialist credit providers.

s160
00:15:46.120 --> 00:15:50.000
And so the way we're seeing it right now is that every deal is bespoke.

s161
00:15:50.000 --> 00:15:51.800
There's different spec sheets, there's formats.

s162
00:15:51.800 --> 00:16:02.440
The bill of materials may not match the actual like delivery or there might be subject to change if a part is missing or part is out of stock and you know, there's like very long lead times

s163
00:16:02.440 --> 00:16:06.640
that some of these OEM and ODM kind of providers have.

s164
00:16:06.640 --> 00:16:09.440
And the same term can mean different things when you're negotiating very quickly.

s165
00:16:09.440 --> 00:16:16.600
If someone's talking about memory, are we talking about you know, VRAM or we talking about the CPU memory, talking about additional storage here?

s166
00:16:16.600 --> 00:16:18.600
Like what did you exactly mean there?

s167
00:16:18.600 --> 00:16:25.800
It affects quite a bit what type of loads can be run on you know, the actual end users needs and also with pricing.

s168
00:16:25.800 --> 00:16:27.240
And delivery dates.

s169
00:16:27.240 --> 00:16:39.280
So there's a lot of back and forth before anyone can really compare and this is where the concept of like non-fungibility kind of comes into place and how to partition that out, what parts can be fungible, what parts hold the most of the weight such as the skew etc.

s170
00:16:39.280 --> 00:16:48.080
And what parts are still like traded as if it was a product, e-commerce etc. where you know, it's only for that particular cluster that this price applies.

s171
00:16:48.760 --> 00:16:50.920
And so prices can update daily or faster.

s172
00:16:50.920 --> 00:16:54.920
Quotes still a good starting point but it can get chopped rather than it is changes what's on offer.

s173
00:16:54.920 --> 00:16:59.920
There's all these kind of latent factors, hidden factors as well why we're calling it a dark market.

s174
00:16:59.920 --> 00:17:04.000
These are some public H100 references snapshotted today.

s175
00:17:04.000 --> 00:17:04.839
I updated them.

s176
00:17:04.839 --> 00:17:09.839
So take a look at the what the public so those are the references for if you wanted to go on their websites to see.

s177
00:17:09.839 --> 00:17:18.160
This is just an example here of like what a public on demand or spot pricing looks like for an H100 today.

s178
00:17:18.160 --> 00:17:20.800
You can see kind of various various prices here.

s179
00:17:20.800 --> 00:17:31.960
And so, people might use this or they might use their own kind of internal what they've seen to actually like assume or use it to charge what they want to charge on their actual cluster specification,

s180
00:17:31.960 --> 00:17:36.520
taking to account every other kind of factor that goes into it.

s181
00:17:36.920 --> 00:17:41.360
And so, why are these new structures kind of needed and what are these microstructures interesting?

s182
00:17:41.360 --> 00:17:43.320
It's mainly to solve for these mismatches, right?

s183
00:17:43.320 --> 00:17:48.040
Bridging capacity, more flexible commitments, resale rights, and protection against price moves.

s184
00:17:48.040 --> 00:17:57.000
Um you know, you're selling a fixed price service from an inference cloud point of view, you'd be selling a fixed price service while the GPU bill floats and that creates price exposure.

s185
00:17:57.000 --> 00:18:04.520
And so, locking capacity for years can solve a part of that, but it'll create idle hour idle hour exposure elsewhere as well, unless they're always using it.

s186
00:18:04.520 --> 00:18:08.520
Reselling spare capacity only helps if someone actually can buy the hours.

s187
00:18:08.520 --> 00:18:15.560
A financial hedge can address a reference price, but can't deliver the cluster right now in terms of cash settlement or guarantee a utilization.

s188
00:18:15.560 --> 00:18:21.560
So, the useful structure has to start with a particular risk that the business has and needs to manage.

s189
00:18:24.320 --> 00:18:29.600
And I'll I'll go pretty fast on this policy section because we're running out of time.

s190
00:18:29.600 --> 00:18:40.680
Everything that a fan has said sits inside of a structure that is regulated and is affected by policy.

s191
00:18:40.800 --> 00:18:42.000
In various ways.

s192
00:18:42.000 --> 00:18:44.960
One way is export controls, one of the examples.

s193
00:18:44.960 --> 00:18:56.080
In January, the US Commerce Department moved qualifying H200 class license reviews for China and Macau to case-by-case review with conditions.

s194
00:18:56.080 --> 00:19:02.000
Furthermore, open weight models connect policy back to the US economics we started with.

s195
00:19:02.000 --> 00:19:14.040
The US AI action plan explicitly encourages open source and open weight AI and recommends improving the financial market for compute so that startups and academics can access large-scale computing

s196
00:19:14.040 --> 00:19:16.040
power.

s197
00:19:17.400 --> 00:19:20.840
So, I'm going to tie this out right here, but where's this opportunity lie?

s198
00:19:20.840 --> 00:19:22.720
We have this growing middle market.

s199
00:19:22.720 --> 00:19:25.240
We need balance sheets that can bridge buyer and supplier needs.

s200
00:19:25.240 --> 00:19:31.880
There needs to be some level of standardization, a standard capacity contract, common specs, common terms, and definitions.

s201
00:19:31.880 --> 00:19:37.240
And so, growing companies with needs and commitment do not fit the contracts the hyperscalers and frontier labs can sign.

s202
00:19:37.240 --> 00:19:39.000
Those companies have real workloads.

s203
00:19:39.000 --> 00:19:40.240
We want them to grow.

s204
00:19:40.240 --> 00:19:43.720
They still face this financing problem, a timing problem, a specification problem.

s205
00:19:43.720 --> 00:19:50.840
A supplier might want the long-term commitment for the build-out while the buyer wants to expand in smaller blocks as demand becomes clearer as it's so volatile, front of curve.

s206
00:19:50.840 --> 00:19:53.800
The buyer needs a way to avoid paying for capacity can't use yet.

s207
00:19:53.800 --> 00:20:02.680
And so, participants with strong balance sheets that are uncorrelated with the market and the ability to assess those risks with a strong credit profile can help bridge the gap between what

s208
00:20:02.680 --> 00:20:04.720
buyers need and what suppliers offers.

s209
00:20:04.720 --> 00:20:07.040
That requires more than quoting the actual GPU hour.

s210
00:20:07.040 --> 00:20:09.560
It means understanding the delivery, workload, all of that.

s211
00:20:09.560 --> 00:20:19.240
And so, there's very few kind of players in the market, and you know, a very well-suited player can be a firm that's uncorrelated with market, with strong balance sheet, strong credit rating, an incredible

s212
00:20:19.240 --> 00:20:24.000
amounts of hardware talent that sits inside for high-performance computing.

s213
00:20:24.000 --> 00:20:27.160
And so, just making a CTA here, let's build this together.

s214
00:20:27.160 --> 00:20:31.720
Thank you guys so much for attending our talk.

s215
00:20:32.724 --> 00:20:34.724
[applause]